The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes files in user-specified directories and includes their content or metadata in a markdown report. Malicious instructions hidden within those files could be included in the report and potentially influence the agent during the suggested 'automatic repair' phase or future interactions.
Ingestion points: The target directory scanned by scripts/entropy-scan.sh.
Boundary markers: The generated report does not use specific delimiters to isolate potentially untrusted content extracted from files.
Capability inventory: The skill uses find and grep to read file system data and writes findings to the ~/clawd/tmp/ directory.
Sanitization: There is no evidence of escaping or filtering content extracted from the target files before it is presented in the report.

entropy-manager